Why we should be intentional about the mental models we use for thinking when we think about digital sustainability

To make sense of our world, we often form incomplete, yet still useful mental models of how it works. Most of the time these are helpful, but when we are dealing with complex systems, what might feel intuitively right, can lead to outcomes totally at odds with we were initially aiming to achieve. This post outlines how this applies when we talk about the environmental impact of digital services.

It’s easy to misapply mental models when faced with complicated systems like the internet, or ‘below’ it, the electricity grid it runs on. To be more specific, it’s easy to form an intuitively attractive mental model in cases like this, that can end up being incorrect in ways that undermine what you were trying to achieve in the first place.

If you work with software, you be familiar with term leaky abstraction – it’s somewhat similar, and a useful concept.

Leaky abstractions usually borrow ideas from another domain we’re already familiar with, to help us make sense of a new one, providing a mental model that for the most part, is helpful. They make working with something new feel like working with something familiar, which often means techniques you already know can be brought to bear – making us more productive, helping us feel more confident when faced with problems, and so on.

One example might be the way programmers use an Object Relational Mapper (ORM) code to make interacting with say… an SQL database feel like working with simpler datastructures, in their preferred language. These are common – in Python, Django’s ORM is well liked, and in the Ruby on Rails community you have ActiveRecord. In Javascript’s Prism.js is popular, and so on. That’s the abstraction part.

The leaky part refers to how an abstraction can fail, or cause a system to work in very different ways to the mental model you had internalised. In an ORM, you might try fetching data in a way that might have felt simple, yet results in an executing a set of incredibly expensive database queries that slow a programme to crawl, or even exhaust the resources on a machine, crashing the server. When this happens, the same model that initially served us well, can turn out to have some significant shortcomings.

This brings us back to models. Or rather, one mental model I see used all the time when people start thinking about how to manage the environmental impact of digital services.

An example model: applying a “cars and driving” consumption model to the resource footprint of computing

When many of us start trying to tackle the environmental footprint of computing, because we’re often billed by companies based on the quantity of something we purchase (i.e. a volumetric basis), it’s easy to reach for similar volumetric models we’ved used successfully elsewhere, to help us think about how pollution might be caused.

For the most part, we don’t see the pollution ourselves on a short time scale when using digital services. In this scenario, when grasping around for a mental to help think about it, you might apply a “cars and driving” model.

Broadly speaking, the more you drive, the more pollution you cause, because when driving in a vehicle that runs on petrol or diesel, a large part of the pollution happens in or around the car, as the fuels are burned and the exhaust gases leave the tailpipe. Regulation in many parts of the world even forces automobile makers to report along these lines, in terms of grams of CO2 per kilometre travelled.

So, if you wanted to reduce the pollution from driving, one obvious step you might take would be to drive less, because driving less mostly means burning less fuel.

What about the other pollution from driving cars?

In this scenario, we’re less likely to think about the other drivers of pollution like the embodied energy in making the car, or even the pollution caused by constructing and maintaining the roads themselves. Even though we know they’re not zero, they don’t change frequently enough in response to our own use to help us form a similarly intuitive model, so the “more driving means more pollution” idea takes precedence.

Where this model can seem a good fit at first: consumption-based billing in cloud and telecoms

As I said before, when thinking about the pollution associated with digital services, this consumption based model is attractive. Crucially because things like cloud computing or data transfer with telecoms are often billed based on the quantity we use, it’s easier to come up with numbers we appear to have some meaningful control of.

Even when we use personal hardware, because we have batteries that fill and empty on timescale of hours, it’s easy to end up applying a usage based model there too, thinking “more usage equals more pollution” once again.

Where this model doesn’t serve us so well, and why we might use it anyway

The problem here is that you can end up with overly simplistic conclusions.

One of the most maligned examples is the idea that if you halve the data sent over a network (by designing a website or reducing the resolution of a video), you halve the emissions caused.

This is compounded by the fact that we already have lots of tools that make it easy to measure things like data sent over the wire, or how much you’re billed each day. These follow the “more usage equals more pollution” model, but just as importantly, they provide a really satisying sense of sense of agency! You feel like you’re really effective when you make changes to an application and you see a number go down, just like you get a lovely warm sense of validation you feel when you see a wall of passing tests.

I think this agency part is really important thing to acknowledge when faced with a big, scary, systemic problem like climate change.

More specifically, I think this because one of the reasons that climate change is such an emotionally draining topic is that it’s psychologically exhausting to know that:

  1. a bad thing is happening
  2. there are no obvious levers you have available to affect it.

In this context, something that seems to give you back some of agency offers immmediate emotional relief, even if you might have some nagging doubts about whether it’s really the best way to think about a problem.

Conversely, the prospect of letting go of a “car and driving” model can be really unattractive. If we do want to reduce the emissions associated with our use of digital services, going back to “driving blind” and losing this sense of agency crave is really unattractive.

When we don’t acknowledge that this sense of relief is a key factor affecting how we think about digital sustainability, I think it hinders our ability to come up with mental models that inform more effective interventions.

This isn’t the only mental we can use though.

A different model: instead of “cars and driving”, “bikes and bike lanes”

Part of the agency problem I think comes down to us being deprived of tools we can use to interrogate a problem, and this is one of the problems associated with rejecting the “cars and driving” model. So what can we do?

One option is to look for other models that might also be useful, are still somewhat familiar, that give clues to other potential interventions, but don’t have such unwanted downsides.

In this context, rather than thinking about “cars and driving”, when I think about digital connectivity specifically and the related pollution, I try to think of “bikes and bike lanes” instead now.

I’ll try unpacking why.

1. There’s still an environmental footprint from making bikes, and making bike lanes

Most of us, if we don’t own or ride a bike, have at least seen a bike, and we have some idea that they’re made from things like metal or rubber, that took energy to create and had some enviromental footprint associated with their production.

Similarly, if we think about the the pollution caused when creating a bike lane – we can picture various forms of roadworks creating it. I think of crews digging up new paths and so on, and even when they’re just painted onto the street, applying durable paint still can be a smelly, polluting process.

2. Once they’re in place, the pollution from riding a bike down a bike lane doesn’t scale linearly with use

Once we have bike lane built, using the bike lane has some environmental footprint, sure, because we need to eat food to turn to the energy we use to pedal. Also, if absolutely loads of us used bike lanes, at some point we’d saturate the bike lane with riders, and we might end building more bike lanes to keep up with demand, which would have its own environmental consequences.

If you really wanted to get granular, you might look at the use phase energy of the streetlights overhead along a bike lane, and attribute responsibility to the riders as they ride under them – doing this would make your numbers go down as you reduce the number of people using the lane, for example, and offer a really clear intervention you could try.

However, for the most part, the difference between one person using a bike lane each morning, and say 100 people using it that morning isn’t that large – and as a share of the total pollution caused by building out the bike lane, it would be tiny.

3. The value offered by bike lanes comes from providing a more convenient, less polluting alternative to something that would otherwise cause more pollution

Finally, we use mostly bike lanes because we want to connect distant places, where the default is often more polluting, like driving in a car, or in some cases public transit.

It might be the case that once someone has travelled somewhere they still carry out activity that has environmental consequences – I might cycle to a restaurant and eat a meal from really carbon intensive ingredients, or do something that uses loads of energy once I’m there for example.

But it would be quite strange to attribute that to the bike lane.

In my view, this isn’t a million miles away from how we use collaboration tools to work with distant people. I know there’s an environmental impact from me having a zoom call for example, and I can even use a zoom call to plan and kick off actions stuff that again have significant environmental consequences.

But for the most part, if I drove all the way to meet someone to do the same thing there would be a greater carbon footprint, even after accounting for how the convenience of jumping on a zoom call with a co-conspirator makes the it happen more common.

4. It doesn’t prescribe too much. There is still scope to make to biking more environmentally friendly without discarding the key benefit bike lanes provide.

If I still wanted to reduce the environmental impact of bike lanes, I would not be robbed of agency in this scenario. I can still apply things I know about bikes and bike lanes to help me think of different interventions.

If I cared about the embodied energy of the bike I ride, rather than buying a brand new bike, I might use a shared mobility service instead. Using a hired bike, where the embodied carbon is shared across all its users over its lifetime would be a win along these dimensions.

If owned a bike, I might choose to buy a bike made with greener materials instead. If cared about the fuel burned, I might think about where the energy comes from, eating more plants and less beef.

If I cared about the impact of making and maintaining the lanes, once again, there’s scope to think about the choice of materials (see my earlier post about zero carbon cement for construction as an example). I might not be able to impact this directly as as consumer, but there are still options to me as a member of society. I could engage with other organisations who do have more direct access to this lever, like local councils, transport groups, lawmakers and so on.

Of course, if I insisted on using “consumption” of the bike lane as my lever for change, I might also choose not to use the bike lane at all. Once again, I could probably come up with a model to show some measurable change in environmental impact by doing so too. It would be at the cost of not being able to see people I care about as often though, and forgoing a bunch of other benefits.

This is also an incomplete model, obvs

There are obviously loads of ways this model falls short as a way to think about digital sustainability in the broader sense, and I’m mainly thinking about digital connectivity in the form of networking when sharing it.

When I think about something like AI or cloud computing, I don’t think this model works that well either.

That said, I’m wary about saying the “car and driving” model makes more sense there, because once again, I think it’s too easy to conflate the way we are billed for compute paid for by the hour with how emissions are caused as a byproduct of making these kind of computing services available like a utility.

I’d really appreciate suggestions for better models, to help think about the problem though.

Summary – all models are wrong, some are useful, and knowing when they stop being useful is important. Also, bikes are nice.

Until we have a well developed sense of intuition about what is helpful, and what isn’t in the world of digital, I think we’ll end up reaching for models we are already familiar with when trying to make sense of this domain.

I hope this helps highlight the importance of being intentional about the mental models we end up using, and how doing so can help point us to interventions that are effective, and also meet a deeper emotional need for agency when faced with big, seemingly intractable problems related to climate change, environmental justice and so on.